You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
niinii (knee-knee) is a graphical frontend for glossing Japanese text. Useful
9
-
for assisted reading of text for language learning purposes. A primary use case
10
-
is glossing visual novels, which is shown in the demonstration above. I made
11
-
this tool with the express intent to read a single *specific* visual novel,
12
-
which is also where the name comes from. If someone else finds it useful that's
13
-
cool too.
8
+
niinii (knee-knee) is a graphical frontend for glossing and translating Japanese
9
+
text. Useful for assisted reading of text for language learning purposes. A
10
+
primary use case is glossing visual novels, which is shown in the demonstration
11
+
above. I made this tool with the express intent to read a single *specific*
12
+
visual novel, which is also where the name comes from. If someone else finds it
13
+
useful that's cool too.
14
14
15
15
For example, in the demonstration above, I use niinii along with a text hooker
16
16
to gloss the dialogue in a visual novel. The segmented phrase along with ruby
17
17
text (i.e. furigana) is displayed. Hovering over a segment will show dictionary
18
18
definitions and inflections from JMDict. You can pop open a separate window by
19
19
clicking on a segment. Hovering over kanji will show kanji information from
20
-
KANJIDIC2. I would write a more detailed user manual but I think you can
21
-
probably figure it out.
20
+
KANJIDIC2. Behold the LLM-based live translation. I would write a more detailed
21
+
user manual but I think you can probably figure it out.
22
22
23
-
Japanese language support is implemented using
23
+
Segmentation is implemented using
24
24
[Ichiran](https://github.com/tshatrov/ichiran) by
25
25
[tshatrov](https://github.com/tshatrov). Ichiran is pretty amazing at text
26
26
segmentation compared to other tools I've tried.
27
27
28
+
As of late, niinii incorporates translation features via LLM offerings from
29
+
OpenAI. Since I started working on this project, the state of LLM-based
30
+
translation has drastically improved both in quality and latency. I won't
31
+
comment on the state of the art here since it changes on a monthly cadence.
32
+
Again, I would write a more detailed user manual but you'll figure it out.
33
+
Bring your own API key.
34
+
28
35
## Why not use...
29
36
This is a tool created to service a personal need, and may not be useful to you.
30
37
Below, I laid out my personal justification for investing time into creating
@@ -36,8 +43,10 @@ much better at segmentation, provides more metadata, and makes fewer mistakes.
36
43
**Why not use rikai(kun|chan), JGlossator?**: They don't do segmentation.
37
44
38
45
**Why not use DeepL, Google Translate, etc.?**: I want a gloss, not a
39
-
translation tool. If I ever integrate translation features, I'd like to do so in
40
-
a way that supplements the gloss rather than dumping text.
46
+
translation tool. ~~If I ever integrate translation features, I'd like to do so in
47
+
a way that supplements the gloss rather than dumping text.~~ I have finally
48
+
incorporated translation features into niinii. Of course, the purpose here is
49
+
still to supplement the gloss.
41
50
42
51
**Why not use the web frontend [ichi.moe](https://ichi.moe)?**:
43
52
There are some features I'd like to experiment with to improve the glossing
@@ -46,7 +55,7 @@ experience.
46
55
## Build
47
56
Prepackaged builds are available in the
48
57
[Releases](https://github.com/Netdex/niinii/releases) section of this
49
-
repository.
58
+
repository. These releases are seldom published and may be missing many unstable features. You may download bleeding-edge builds from [GitHub Actions](https://github.com/Netdex/niinii/actions).
50
59
51
60
The only target that is properly maintained is `x86_64-pc-windows-msvc`. There's
52
61
nothing stopping it from working on other targets (e.g. Linux), but additional
0 commit comments